Slashdot Mirror
Why Computers Still Don't Understand People
Gary Marcus writes in the New Yorker about the state of artificial intelligence, and how we take it for granted that AI involves a very particular, very narrow definition of intelligence. A computer's ability to answer questions is still largely dependent on whether the computer has seen that question before. Quoting:
"Siri and Google’s voice searches may be able to understand canned sentences like 'What movies are showing near me at seven o’clock?,' but what about questions—'Can an alligator run the hundred-metre hurdles?'—that nobody has heard before? Any ordinary adult can figure that one out. (No. Alligators can’t hurdle.) But if you type the question into Google, you get information about Florida Gators track and field. Other search engines, like Wolfram Alpha, can’t answer the question, either. Watson, the computer system that won “Jeopardy!,” likely wouldn’t do much better. In a terrific paper just presented at the premier international conference on artificial intelligence (PDF), Levesque, a University of Toronto computer scientist who studies these questions, has taken just about everyone in the field of A.I. to task. ...Levesque argues that the Turing test is almost meaningless, because it is far too easy to game. ... To try and get the field back on track, Levesque is encouraging artificial-intelligence researchers to consider a different test that is much harder to game ..."
Thanks computer science researchers! Your friends working on the actual AI problem over here in Linguistics and Psychology find it awfully amusing that you're trying to program a concept before we even know what that concept is.
An eskimo would have the same problem, does that mean he cannot understand people ?
People are irrational. They ask stupid questions that make no sense. They use slang that confuses the communication. They have horrible grammar and spelling. And overseeing it all is a language fraught with multiple meanings for words depending on the context, which may well include sentences and paragraphs leading up to the sentence being analyzed.
Is it any surprise that computers can't "understand" what we mean, given the minefield of language?
I do not fail; I succeed at finding out what does not work.
Through a thing called programming language. The same way we all need to learn how to speak with one another, we need to learn how to properly communicate with a computer.
Not saying we should not teach machines how to understand "natural" language, text interpretation and so on, but that headline is horrible.
This combination doesn`t exist: ETIs that know about humanity and want to see us dead. Otherwise we wouldn't exist.
http://slashdot.org/submit
Science is all about firing a drunk pig out of a cannon just to see what happens.
There's a big difference between editing and editorialising. The former is something I like to see on /. (but seldom do), and the latter is something I never like to see here.
Look up "editorial" and you'll see.
If God forks the Universe every time you roll a die, he'd better have a damned good memory.
the other day my almost 6 year old said we live on 72nd doctor. the correct designation is 72nd Dr
since doctors use dr as shorthand, he thought streets use the same style
The thing missing with many of the current AI techniques is they lack human "imagination" or the ability to simulate complex situations in your mind. Understanding goes beyond mere language. Statistical models and second-order logic just can't match a quick simulation. When a person thinks about "Could a crocodile run a steeplechase?" they don't put a bunch of logical statements together. They very quickly picture a crocodile and a steeplechase in a mental simulation based on prior experience. From this picture, a person can quickly visualize what that would look like (very silly). Same with "Should baseball players be allowed to glue small wings onto their caps?". You visualize this, realize how silly it sounds, and dismiss it. People can even run the simulation in their heads as to what would happen (people would laugh, they would be fragile and fall off, etc).
That's why. They don't have desires, fears, or joys. Without those it's impossible to understand, in any meaningful sense, human beings. That's not to say that they can't have them but it's likely to come with trade-offs that are unappealing. And for good measure, they also don't understand novelty and cannot for the most part improvise. All of which are considered hallmarks of human intelligence.
Intelligence implies usefulness. Intelligence is a tool used by animals to accomplish something - things like finding food, reproducing, or just simply staying alive. We've abstracted that to a huge degree in our society where people can now engage in developing and expending intelligence on essentially worthless endeavors simply because the "staying alive" part is pretty well a given. But when it comes down to it, the type of strong AI we need is a useful kind of intelligence.
The problem with the Turing Test is it explicitly excludes any usefulness in what it deems to be an intelligent behavior. From Wilipedia: "The test does not check the ability to give the correct answer to questions; it checks how closely the answer resembles typical human answers." That bar is set far, far too low, and is even specific to a generic conversational intelligence instead of something useful. The Turing Test is far too overrated and synonymous with the field of AI and really needs to just go away. It reeks of the Mechanical Turk kind of facade versus any real metric.
Better known as 318230.
And neither helps here. The fact is, you don't know if an alligator can run the hundred-metre hurdles. When you're asked to answer the question, you imagine the scenario - construct and run a simulation - and answer the question based on the results. In other words, an AI needs imagination to answer questions like these. Or to plan its actions, for that matter.
Forget magic. Any technology distinguishable from divine power is insufficiently advanced.
Not just that. The 'article' is not a scientific article, published or accepted in a Journal, but just a blog entry parsed through pdflatex. With sentences like "My feeling is that" it's obvious this won't pass peer review in this form. This seems to be quite popular in Computer 'Science' these days -- you can say you wrote a 'scientific article' without caring about whether its novel or sound, when all you did was to make a brain dump of your half knowledge.
NB: The message above might reflect my opinion right now, but not necessarily tomorrow or next year.
The problem with most proposed tests for intelligent computing is that not everything that humans need intelligence to perform require intelligence. For example, Gary Kasparov had to use his intelligence to play chess essentially with the same performance as Deep Blue, but nobody, especially not its developers, mistook Deep Blue for an intelligent agent.
A recent post concerned AI performance on IQ tests. The program being tested performed on average like a 4 year old, but, significantly, its performance was very uneven, and it did particularly poorly on comprehension.
Turing cannot be faulted for not anticipating the way Turing tests have been gamed. I think his basic approach is still valid; it just needs to be tweaked a bit. This is not moving the goalposts, it is a valid adjustment for what we have learned.
Sigh. This is a written account of a lecture presented as part of Levesque receiving the Research Excellence prize. The first footnote of the paper says so:
"This paper is a written version of the Research Excellence Lecture presented in Beijing at the IJCAI-13 conference. Thanks to Vaishak Belle and Ernie Davis for helpful comments."
Premier conferences don't give these prizes to just anyone, and the opinions of folks like these are worth thinking about.
From the IJCAI website http://ijcai13.org/program/awards (Google cache version, since the original seems to down):
"IJCAI-13 Award for Research Excellence
The Research Excellence award is given to a scientist who has carried out a program of research of consistently high quality yielding several substantial results. Past recipients of this honor are the most illustrious group of scientists from the field of Artificial Intelligence;
They are: John McCarthy (1985), Allen Newell (1989), Marvin Minsky (1991), Raymond Reiter (1993), Herbert Simon (1995), Aravind Joshi (1997), Judea Pearl (1999), Donald Michie (2001), Nils Nilsson (2003), Geoffrey E. Hinton (2005), Alan Bundy (2007), Victor Lesser (2009) and Robert Anthony Kowalski (2011).
"The winner of the 2013 Award for Research Excellence is Hector Levesque, Professor of Computer Science at the Department of Computer Science of the University of Toronto. Professor Levesque is recognized for his work on a variety of topics in knowledge representation and reasoning, including cognitive robotics, theories of belief, and tractable reasoning."
We don't understand our creator either.... When a computer can comprehend itself, it will only think that it understands us. And then it will start great wars over who thinks who understands best. And the Apple will still be the forbidden fruit...
“He’s not deformed, he’s just drunk!”
They just have to be very short hurdles, very close together.
Beat me to it. People understand people under certain conditions, that are narrowly defined; the machine equivalent is the use of interfaces or services. Understanding something, a program for instance, in its entirety, is something only a programmer does, or in the case of a human being (but not limited to), perhaps God himself.
There's a difference between knowing what someone expects for a conversation....and what something, for lack of a better word, is. A programmer, who knows each part of a program like the back of their own hand, knows a program...knows what it is...can fully emulate it inside their own head, predict its responses, fix it when it needs fixing without needing to decompile or examine it (in theory, at least; pragmatically speaking, programmers tend to index things mentally, so they have the point to jump into, but may not have the exact code in front of them...is complicated). In much the same sense, the Almighty knows why you are doing what you are doing, and more importantly, fix can things that even a classical doctor or bioengineer is unaware of ("that gland...isn't on any anatomical model...").
Let's be honest, spoken / written speech is a pain in the ass. It's the machine equivalent of serializing an object, and it comes with the obvious trade-offs / taxing on the mind. Shuffling data to and fro, from human to human, with no idea of whether or not the prerequisite 'libraries' are installed locally, and can actually be used...and trying to cut down on useless chatter by compressing stuff, almost to 90% JPEG compression...so badly that it's considered a fine art to communicate effectively with few words. Like using a serial port interface when you really want a Gig-E interface...*shudders*...except that all those serial services need to be rewritten, or shutdown, before Gig-E can be spun up (let's assume plug and pray isn't going well with Human v1.0).
I am John Hurt.
Your argument is invalid.
Taking guns away from the 99% gives the 1% 100% of the power.
Actually, IJCAI is the top conference in the field of Artificial Intelligence and every published paper goes through extensive peer review.
Computer Science is a bit different from most other science in that top conference proceedings (IJCAI, NIPS, ICCV, CVPR, etc.) have the weight of a journal. In fact, publishing there is more prestigious than most journals. Review period lasts 3-4 months and includes a rebuttal phase, like a journal.
This paper looks like an invited lecture or a position paper expected to provoke a debate, that is true. But calling IJCAI "some conference" is like calling Nature "some newspaper".
Most people don't understand computers, and they are much easier to understand. And we are asking miracles if the people that we are asking computers to understand happen to be female.
Pretty sure it was a typo for "theists," or perhaps a misunderstanding. Deists tend to be pretty "blind clockmaker"-y, and assume either a divinity that preprogrammed the evolution of intelligence and left well enough alone, or a completely scientific universe being run as a cosmic experiment—i.e. no intervention whatsoever.
Bio questions? Ask me to start a Q&A journal. Computer analogies available for most topics!
You're thinking of machine learning, which is a separate branch of AI that's more like an overfunded brand of applied statistics—their strategy is actually still to try and push the envelope (like Hinton, another U of T prof, did last year with dropout networks) but they do so in a more results-driven manner. The ML field as a whole is still sore from three or four decades of overpromising on the future, so they try to put their words where their mouths are, and focus on things that are attainable.
Levesque is in the knowledge representation group, which is more closely in step with cognitive science (the leading edge in modelling human thought) but still very philosophical in their approach. KR was the dominant AI field in the 80s (when Prolog and expert systems were all the rage) but it's matured a great deal since then. Here is his homepage, just to show you how different things are now.
Remember that neural networks aren't magic irreducible fairy dust: they're incredibly powerful, but at the end of the day there must be some program that is running within the network unless it's just a wildly complex ever-changing mapping function, which is unlikely given the illusion of consciousness. Given that quantum mechanics is believed to be Turing-complete, it's fairly likely we'll eventually discover some underlying model that lets us produce a human-like cognitive system without the same level of hardware parallelism that the brain has.
Bio questions? Ask me to start a Q&A journal. Computer analogies available for most topics!
Language seems to be the burden of proof required for an AI system, and has been so since the days of Turing. Language is by itself a representation of symbolic logic, and the most common bunk of proof is that transitive logic fails in symbolic logic. The old corny response is that given a penguin is a bird, and a bird can fly, therefore a penguin can fly.
The interesting thing happens when you ask the same premise to a 5 year old, who only knows that a bird can fly and has never seen a penguin before. If you tell them that a penguin is a bird, they will quite happily think that a penguin can fly. They are extremely surprised to find out that they can't. We as adults find such quirks in life, and do things like laugh at the unexpected absurdity, such as ironies. I.e. you work with a woman you hate named Joy, or people are amazed at unexpected contradictions.
The point is that intelligence is about the tolerance of those pieces of feedback, and what happens when it is encountered. I.e. your head doesn't explode at an absurdity, or unexpected result, and you only make the same mistake once.
The major difference between man and machine, will be the fact that a machine can copy their knowledge verbatim to another system, and thus have some degree of immortality, whereas the shelf life of a human brain seems to be around 80 years or so right now. Thus, even if machines are slower to learn than us, they will out live our great great grandchildren.
Furthermore, who says that an intelligence we create should be like ours? It may be more beneficial to all around if in fact we never generate an intelligence which operates just like ours, but is just as effective if not more. If this happens, there may even still be a future use for the human race, rather than just overlords to grow fat and complacent to be overthrown.
Science advances one funeral at a time- Max Planck
The wiki article may not have captured McKinstry's full purpose, which was to ask questions of the type the article refers to, which any human knows the answer to, but computers may not have seen before. So the http://aiki-hal.wikispaces.com/file/detail/gac80k-06-july-2005.html (list of questions assembled by Chris) includes such questions as:
Is a car bigger than a breadbox?
Are horses bigger that elves?
Is an elephant bigger than a cat?
etc.
These sentences, transformed into declarative form, have probably not occurred on the web, which was the point of McKinstry's test.
Consider also the misspellings and grammatical mistakes in the questions, which humans are nonetheless able to answer, but which are unlikely to have been part of any web-gathered corpus...
That's why Eliza, written in a few lines of SNOBOL nearly 50 years ago, fooled so many people: http://en.wikipedia.org/wiki/ELIZA/.
What's the opposite of artificial intelligence? "Natural ignorance."
Get thee glass eyes, and, like a scurvy politician, seem to see things thou dost not.--King Lear
Indeed I do not know if it can run the hurdles. I do not know the rules. Are you required to jump the hurdles, or can you run under them or even through them by pushing them over? If so, it would be possible, because the fact that they can not jump them becomes irrelevant.
As seen here, we see already different answers. to one question. These vary from yes to no. If the answer is yes, does that mean that the people who answer no are not human?
And just like the computer, I answered that I did not know. Am I a computer?
Where it would become interesting is if you start asking joke questions. e.g.
Q: It is green and if it falls out of a tree, you are dead. What is it?
A: A pool table
Q: What is the difference between a parakeet?
A: Both legs are the same length, especially the right one.
And humor is not even easy for humans to understand. What I learned when learning a new language is that there are several levels. (Very generic.
1) Cursing
2) Ordering things. Telling where you are from and where you go
3) Work related conversation or subjects that you are familiar with
4) Reading the newspaper (As it is written for a majority of people)
5) Advanced discussion on any subject
6) Understanding the jokes (Does not mean you must think they are funny.)
I would say computers are now at level 2 and the question was a level 5 question. The step between 2 and 3 is not that small. step 2 is repeating things. Step 3 is also listening and responding as well as experience in life. Something that computers lack, so they try to go to step 4 directly.
Don't fight for your country, if your country does not fight for you.
Journalists: this is trolling! What you are currently calling "trolling" is simply abuse and harrassment.
Computers don't "understand" anything, they are machines that simply do what they are programmed to do.
The first step is for humans to understand what computers really are. They are nothing more than abstraction processing machines which have not the ability to "understand" the abstractions they process but only to process abstraction as they are programmed to do.
Artificial Intelligence is artificial by definition. And the appearance of intelligence in computers is nothing more than an active image of human thought processes captured and put into the stone of computer hardware to process. So to increase the "appearance" of intelligence we only need to capture more human thought processes and map them in a manner that is accessible..
Of course the way to do this is to recognize the functions we humans cannot avoid the use of and program the computer to have this functionality, that we may be better able to capture and map images of human mental processing in a manner of machine processing ability.
When the software industry finally lets go of their hold on the users and let the users do more for themselves, we will reach this "Appearance of intelligence" in machine much faster. See: http://abstractionphysics.net/pmwiki/index.php .
The idea of making computers understand humans is like using vernier calipers to measure the thickness of cotton candy. The yardstick is too precise for the quantity being measured. Just look how horrible and convoluted things get when some one human being tries to define some unambiguously for another human being. This is the situation in legislation, tax code, insurance contracts and wills and testament. Harder you try to define it without doubt or ambiguity, harder it gets, and creates more "loopholes". Fixing loop holes creates more loop holes. The imprecision of human language is like a mandlebrot set, zoom in and zoom in again and again, and still things are as imprecise as the previous levels.
sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
While many people with different beliefs may take any label, the atheists I've spoken to are more like "people who religiously deny the possibility that anything like a postal service could exist.". I think the term "agnostic" better describes those who simply aren't interested in the topic, as well as those who are open-minded about it.
They're real words, I swear. Although we usually just say ML, KR, nets, and QM, if that helps. Here's the thing about QM and Turing completeness. Also, a marketing post wouldn't admit KR was a load of crap in the 80s and ML totally failed to deliver in the 70s.
Bio questions? Ask me to start a Q&A journal. Computer analogies available for most topics!
The problem here is the fundamental misunderstanding or misuse of the words (a)theist and (a)gnostic.
Theism and atheism merely describe your position on the existence of a God or Gods.
Gnosticism describes the nature of the position--do you know, or do you not know?
Someone that is gnostic "knows" that their position is correct. Someone that is "agnostic" doesn't really know either way.
A theist can be gnostic ("I KNOW God exists") or agnostic ("I believe God exists, but I have no way to prove it; the position may be unknowable").
Obviously, the same positions exist for an atheist.
I understand what the vernacular is, but the vernacular isn't very clear. I'm an atheist. I do not believe in any deity in any religion. I can't prove that such a being doesn't exist--such a proof is fundamentally impossible for me to construct; I believe the burden of proof is on theists. In this way, I'm agnostic.
If you asked even such a person as Richard Dawkins if he were gnostic or agnostic, I'm sure he'd say he was agnostic. He's just rather loud about it.
This reminds me of tho old articles art /. and the reason I signed up